This is the programming code for case study one. The code includes the experiments for both SHAP and LIME explaining tabular data processing problems based on four machine learning methods. The four methods are (a) Random Forest, (b) Logistic Regression, (c) XGBoost, and (d) K-nearest neighbours (KNN).

from __future__ import print_function
import time
import sklearn
import sklearn.datasets
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LogisticRegression
from xgboost import XGBClassifier
from sklearn.neighbors import KNeighborsClassifier
import numpy as np
import lime
import lime.lime_tabular
import shap
#np.random.seed(1)

LIME

Random forest

# run the machine learning model
X = sklearn.datasets.load_iris()
train,test,train_label,test_label = sklearn.model_selection.train_test_split(X.data,X.target,train_size=0.80)
model_rf = RandomForestClassifier(n_estimators=500)
model_rf.fit(train,train_label)
print(sklearn.metrics.accuracy_score(test_label, model_rf.predict(test)))

0.8666666666666667

# generate the LIME tabular explainer
start = time.clock()
lime_explainer = lime.lime_tabular.LimeTabularExplainer(train,
                                                   feature_names=X.feature_names,
                                                   class_names=X.target_names,
                                                   discretize_continuous=True)
end = time.clock()
# calculate the runtime
print('The function run time is: ',end-start,' second(s)')

The function run time is:  0.007305700000003412  second(s)

# explain a single prediction
i = np.random.randint(0, test.shape[0])
print(i)
j = np.random.randint(0, test.shape[0])
start = time.clock()
# explain the first case
exp1_rf = lime_explainer.explain_instance(test[i], model_rf.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# explain the second case
exp2_rf = lime_explainer.explain_instance(test[j], model_rf.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')

4
The function run time is:  2.6685693000000015  second(s)
The function run time is:  2.507269600000029  second(s)

# show the explanation results
exp1_rf.show_in_notebook(show_table=True,show_all=True)
exp2_rf.show_in_notebook(show_table=True,show_all=True)

Logistic Regression

# run the machine learning model
X = sklearn.datasets.load_iris()
train,test,train_label,test_label = sklearn.model_selection.train_test_split(X.data,X.target,train_size=0.80)
model_lr = LogisticRegression()
model_lr.fit(train,train_label)
print(sklearn.metrics.accuracy_score(test_label, model_lr.predict(test)))

0.9666666666666667

# generate the LIME tabular explainer.
start = time.clock()
lime_explainer = lime.lime_tabular.LimeTabularExplainer(train,
                                                   feature_names=X.feature_names,
                                                   class_names=X.target_names,
                                                   discretize_continuous=True)
end = time.clock()
# calculate the runtime
print('The function run time is: ',end-start,' second(s)')

The function run time is:  0.007947500000000218  second(s)

# explain a single prediction

# explain the first case
start = time.clock()
exp1_lr = lime_explainer.explain_instance(test[i], model_lr.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# explain the second case
exp2_lr = lime_explainer.explain_instance(test[j], model_lr.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')

14
The function run time is:  2.334795799999995  second(s)
The function run time is:  2.178248499999995  second(s)

# show the explanation results
exp1_lr.show_in_notebook(show_table=True,show_all=True)
exp2_lr.show_in_notebook(show_table=True,show_all=True)

XGBoost

# run the machine learning model
X = sklearn.datasets.load_iris()
train,test,train_label,test_label = sklearn.model_selection.train_test_split(X.data,X.target,train_size=0.80)
model_xgb = XGBClassifier(eval_metric='mlogloss')
model_xgb.fit(train,train_label)
print(sklearn.metrics.accuracy_score(test_label, model_xgb.predict(test)))

0.9333333333333333

# generate the LIME tabular explainer
start = time.clock()
lime_explainer = lime.lime_tabular.LimeTabularExplainer(train,
                                                   feature_names=X.feature_names,
                                                   class_names=X.target_names,
                                                   discretize_continuous=True)
end = time.clock()
# calculate the runtime
print('The function run time is: ',end-start,' second(s)')

The function run time is:  0.007662999999979547  second(s)

# explain a single prediction

# explain the first case
start = time.clock()
exp1_xgb = lime_explainer.explain_instance(test[i], model_xgb.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# explain the second case
exp2_xgb = lime_explainer.explain_instance(test[j], model_xgb.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')

3
The function run time is:  2.838890700000036  second(s)
The function run time is:  2.247566899999981  second(s)

# show the explanation results
exp1_xgb.show_in_notebook(show_table=True,show_all=True)
exp2_xgb.show_in_notebook(show_table=True,show_all=True)

KNN

# run the machine learning model
X = sklearn.datasets.load_iris()
train,test,train_label,test_label = sklearn.model_selection.train_test_split(X.data,X.target,train_size=0.80)
model_knn = KNeighborsClassifier()
model_knn.fit(train,train_label)
print(sklearn.metrics.accuracy_score(test_label, model_knn.predict(test)))

0.9

# generate the LIME tabular explainer
start = time.clock()
lime_explainer = lime.lime_tabular.LimeTabularExplainer(train,
                                                   feature_names=X.feature_names,
                                                   class_names=X.target_names,
                                                   discretize_continuous=True)
end = time.clock()
# calculate the runtime
print('The function run time is: ',end-start,' second(s)')

The function run time is:  0.006422100000008868  second(s)

# explain a single prediction

# explain the first case
start = time.clock()
exp1_knn = lime_explainer.explain_instance(test[i], model_knn.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# explain the second case
exp2_knn = lime_explainer.explain_instance(test[j], model_knn.predict_proba, num_features=2, top_labels=1)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')

24
The function run time is:  2.519004100000018  second(s)
The function run time is:  2.333920400000011  second(s)

# show the explanation results
exp1_knn.show_in_notebook(show_table=True,show_all=True)
exp2_knn.show_in_notebook(show_table=True,show_all=True)

SHAP

Random Forest

# run the machine learning model
train,test,train_label,test_label = sklearn.model_selection.train_test_split(*shap.datasets.iris(), test_size=0.2, random_state=0)
model_rf = RandomForestClassifier(n_estimators=500)
model_rf.fit(train,train_label)

RandomForestClassifier(n_estimators=500)

# explain single prediction
shap.initjs()
start = time.clock()
# generate SHAP kernel explainer
shap_explainer = shap.KernelExplainer(model_rf.predict_proba,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[12,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[12,:])

Using 120 background data samples could cause slower run times. Consider using shap.sample(data, K) or shap.kmeans(data, K) to summarize the background as K samples.

The function run time is:  0.07750300000043353  second(s)
The function run time is:  0.21299480000016047  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[29,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[29,:])

The function run time is:  0.21647029999985534  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test)

The function run time is:  6.312022699999943  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# generate the SHAP explainer
shap_explainer = shap.Explainer(model_rf.predict,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# calculate the SHAP value
shap_values = shap_explainer(test)
# plot the results
shap.plots.waterfall(shap_values[0],max_display=4)

The function run time is:  0.0015423999975610059  second(s)

shap.plots.beeswarm(shap_values)

shap.plots.bar(shap_values)

Logistic Regression

# run the machine learning model
train,test,train_label,test_label = sklearn.model_selection.train_test_split(*shap.datasets.iris(), test_size=0.2, random_state=0)
model_lr = LogisticRegression()
model_lr.fit(train,train_label)

lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression

LogisticRegression()

start = time.clock()
# generate SHAP kernel explainer
shap_explainer = shap.KernelExplainer(model_lr.predict_proba,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[12,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[12,:])

Using 120 background data samples could cause slower run times. Consider using shap.sample(data, K) or shap.kmeans(data, K) to summarize the background as K samples.

The function run time is:  0.003322599997773068  second(s)
The function run time is:  0.04029330000048503  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[29,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[29,:])

The function run time is:  0.04136749999997846  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test)

The function run time is:  1.3327983999988646  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# generate the SHAP explainer
shap_explainer = shap.Explainer(model_lr.predict,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# calculate the SHAP value
shap_values = shap_explainer(test)
# plot the results
shap.plots.waterfall(shap_values[12],max_display=4)

The function run time is:  0.001964099999895552  second(s)

shap.plots.beeswarm(shap_values)

shap.plots.bar(shap_values)

XGBoost

# run the machine learning model
train,test,train_label,test_label = sklearn.model_selection.train_test_split(*shap.datasets.iris(), test_size=0.2, random_state=0)
model_xgb = XGBClassifier(eval_metric='mlogloss')
model_xgb.fit(train,train_label)

The use of label encoder in XGBClassifier is deprecated and will be removed in a future release. To remove this warning, do the following: 1) Pass option use_label_encoder=False when constructing XGBClassifier object; and 2) Encode your labels (y) as integers starting with 0, i.e. 0, 1, 2, ..., [num_class - 1].

XGBClassifier(base_score=0.5, booster='gbtree', colsample_bylevel=1,
              colsample_bynode=1, colsample_bytree=1, eval_metric='mlogloss',
              gamma=0, gpu_id=-1, importance_type='gain',
              interaction_constraints='', learning_rate=0.300000012,
              max_delta_step=0, max_depth=6, min_child_weight=1, missing=nan,
              monotone_constraints='()', n_estimators=100, n_jobs=4,
              num_parallel_tree=1, objective='multi:softprob', random_state=0,
              reg_alpha=0, reg_lambda=1, scale_pos_weight=None, subsample=1,
              tree_method='exact', validate_parameters=1, verbosity=None)

start = time.clock()
# generate SHAP kernel explainer
shap_explainer = shap.KernelExplainer(model_xgb.predict_proba,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[12,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[12,:])

Using 120 background data samples could cause slower run times. Consider using shap.sample(data, K) or shap.kmeans(data, K) to summarize the background as K samples.

The function run time is:  0.39441440000155126  second(s)
The function run time is:  0.08535140000094543  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[29,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[29,:])

The function run time is:  0.06464950000008685  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test)

The function run time is:  2.784742699997878  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# generate the SHAP explainer
shap_explainer = shap.Explainer(model_xgb.predict,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# calculate the SHAP value
shap_values = shap_explainer(test)
# plot the results
shap.plots.waterfall(shap_values[12],max_display=4)

The function run time is:  0.001089900000806665  second(s)

shap.plots.beeswarm(shap_values)

shap.plots.bar(shap_values)

KNN

# run the machine learning model
train,test,train_label,test_label = sklearn.model_selection.train_test_split(*shap.datasets.iris(), test_size=0.2, random_state=0)
model_knn = KNeighborsClassifier()
model_knn.fit(train,train_label)

KNeighborsClassifier()

start = time.clock()
# generate SHAP kernel explainer
shap_explainer = shap.KernelExplainer(model_knn.predict_proba,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[12,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[12,:])

Using 120 background data samples could cause slower run times. Consider using shap.sample(data, K) or shap.kmeans(data, K) to summarize the background as K samples.

The function run time is:  0.0091552000012598  second(s)
The function run time is:  0.0520312999979069  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test.iloc[29,:])
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test.iloc[29,:])

The function run time is:  0.05405789999895205  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# calculate the SHAP value
shap_values = shap_explainer.shap_values(test)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# plot the results
shap.force_plot(shap_explainer.expected_value[0],shap_values[0],test)

The function run time is:  1.3509833999996772  second(s)

Visualization omitted, Javascript library not loaded!
Have you run `initjs()` in this notebook? If this notebook was from another user you must also trust this notebook (File -> Trust notebook). If you are viewing this notebook on github the Javascript has been stripped for security. If you are using JupyterLab this error is because a JupyterLab extension has not yet been written.

start = time.clock()
# generate the SHAP explainer
shap_explainer = shap.Explainer(model_knn.predict,train)
end = time.clock()
print('The function run time is: ',end-start,' second(s)')
# calculate the SHAP value
shap_values = shap_explainer(test)
# plot the results
shap.plots.waterfall(shap_values[12],max_display=4)

The function run time is:  0.0023669000001973473  second(s)

shap.plots.beeswarm(shap_values)

shap.plots.bar(shap_values)